4) Instagram

To understand Instagram, one can start from a core premise: it does not simply chase the strongest stimuli; rather, it seeks to maintain stability and predictability within the user experience. The goal of the algorithm is not to create as many "viral hits" as possible, but to filter for each user the "accounts and content they are likely to want to see repeatedly." Under this logic, relationship signals carry immense weight. The system makes a comprehensive judgment on whether an effective "attention channel" still exists between you and a specific account. Following is merely the lowest threshold; what truly carries weight is whether interactions and return visits have occurred recently: DMs, mutual likes, mutual comments, repeat viewings, clicking through to a profile, saving, or sharing to a specific person. These actions are not just "engagement"; they are proof to the system that the relationship remains active.

Instagram does not equate the act of following with the establishment of a relationship; it views a "follow" more as a permit. What truly determines distribution priority is whether a continuous, low-friction, non-forced exchange of attention occurs afterward. When this exchange becomes sparse or breaks off, the system naturally reduces distribution intensity. This is not a punishment, but a conservation of resources: the platform will not proactively push content to someone who has long ceased to respond. This explains a common phenomenon: content quality has not significantly declined, yet reach gradually shrinks. The cause is often not the content itself, but the cooling of relationship signals. If an account changes topics, shifts its rhythm, disappears for a while, or reduces two-way interaction, the system interprets this as a decline in the accessibility of the original channel, placing distribution at a lower priority.

Furthermore, Instagram clearly distinguishes between different intensities of interaction. Public likes and casual comments are relatively weak signals, while DMs, saves, shares to specific individuals, and repeat viewings are stronger because they are closer to authentic demand rather than social posturing. The system favors these signals because they better predict future investments of attention. Therefore, when you post, the system first combines existing relationships with interaction history to judge who is most likely to stay, finish watching, or interact again. If these channels are "hot," even mediocre content is more likely to be delivered stably; if the channel has cooled, the system will distribute more cautiously regardless of production quality.

The second layer is identity recognizability. Instagram defaults to the idea that an account corresponds to a long-term subject, not a series of independent experiments. The system constantly tries to answer: "When a user swipes to you, can they quickly judge who you are, what you are generally doing, and why they should continue watching you?" When an account's output undergoes drastic and unpredictable changes, the system does not immediately view this as innovation; instead, it often shows hesitation. In such cases, the algorithm finds it harder to judge whom to recommend you to and harder to predict if you still meet the expectations of your original audience. The result is often a contraction of reach while the system observes the core audience's reaction. "Identity consistency" here does not mean content must be monolithic or repetitive, but that there exists a followable range of change. Growth, expansion, and pivots are permitted, but if the theme, tone, value stance, and visual language fracture simultaneously, both the system and the audience will need to recalibrate. To avoid pushing uncertainty to strangers too early, the algorithm slows down. This is why Instagram is unfriendly to "random personality switching." Frequent jumping between entirely different expressive identities makes it difficult for the recommendation system to decide which category of a user's daily browsing you belong in. In this context, the contraction of distribution is not a rejection of content, but the platform waiting for the account to re-establish a stable signal.

The third layer is temporal continuity. Instagram cares deeply about whether an account presents a rhythm of "a person appearing consistently." Erratic disappearances, sudden high frequency, and then going dark again are interpreted as uncertain signals of relationship maintenance. The system prefers a predictable presence over emotional or episodic outbursts. This does not require high-frequency updates, but rather a predictable cadence. Whether it is once a week or slower, as long as it appears to be a stable output of life, the system finds it easier to maintain distribution. When an account returns after a long hiatus, the algorithm first observes: is this a brief appearance or a true return?

The fourth layer is content feedback, but in Instagram, this is used more to reinforce relationships than to determine virality. The significance of likes, comments, and saves lies not in their quantity, but in whether they come from an existing interaction relationship. Saves, in particular, are viewed as a "private-use signal," representing that the content has functional value in someone's life. When the same group of people interacts with you repeatedly, the system reads not that "this content is exciting," but that "this account is useful in the daily lives of a specific group." This signal is more important than one-time high engagement because it implies sustainability.

While exploration and discovery do exist, they are conservative. Explore and Reels act more as risk-controlled amplifiers than experimental labs. Usually, the system is only willing to introduce an account to strangers if the existing relationships, identity signals, and rhythm are relatively stable. This expansion is more of a reward than a test. Consequently, Instagram is very cautious about "gambler-style" growth. Briefly catering to trends, drastic pivots, or sudden personality shifts—even if they yield short-term data—rarely gain long-term credit. The platform prefers to slowly amplify a predictable account rather than bet on a high-uncertainty flashpoint. Overall, Instagram is not designed for "event-based" content; it serves "daily-based" presence. It repeatedly evaluates not explosive power, but "co-habitability": whether this account can be encountered repeatedly without disrupting the user's rhythm. Here, "being remembered" does not mean being remembered by a mass of strangers, but being judged by the system as having embedded yourself into the cognitive structure of a small subset of people. As long as this holds true, the platform has reason to retain you, maintain you, and slowly expand you as conditions mature. In this sense, Instagram's algorithm is more of a social memory management mechanism than a content competition system. What it constantly does is filter and retain: who deserves to be seen long-term, and who can be gradually faded out. The criterion is not how loud you are, but how stable, clear, and "co-habitable" you can be over time.

5) Red (小红书) vs. Instagram

Instagram is closer to a display-based social network rooted in relationships and identity. Who you are and your interaction history often matter more than the individual piece of content. Content largely serves the function of continuously updating a personal image and maintaining a social presence. While recommendation mechanisms exist—especially reinforced in Explore and Reels—in the core scenarios like the Feed and Stories, interaction history, follow relationships, and identity continuity remain dominant. Overall, it is a network of long-term nodes rather than a high-frequency test lab; the exposure rhythm for creators is usually relatively smooth, with changes occurring more between different entry points rather than sudden platform-wide amplification or erasure.

XiaoHongShu (Red), by contrast, organizes content and distribution clearly around "referential experience." Rather than a content platform, it is more like an experience-based communication system centered on real-world decision-making. Users come here not to be stimulated or entertained, but to judge: Should I do something? How do I choose? Is it worth it? Will I regret it? Under this premise of use, whether the content constitutes a "borrowable experience" for others is often more important than expressive intensity or emotional density.

Treating Red simply as an "expression platform" often leads to misdirected effort. A more accurate understanding is that content here needs to present a path of judgment. Even for emotional content, that which can be repeatedly spread and retained usually implicitly contains specific contexts, prerequisites, and a process of personal interpretation. When reading, users focus not just on the emotion itself, but on whether this experience can help them pre-simulate a judgment. If this transferability is lacking, the content—even if it gains short-term exposure—will struggle to find sustained distribution.

From the perspective of user behavioral feedback, distribution on Red does not only revolve around immediate emotional reactions. Dwell time, repeat viewings, collections (bookmarks), and subsequent click-throughs in search scenarios reflect whether content has been "used" far more accurately than a simple like. These behavioral signals do not necessarily mean the content will go viral instantly, but they point toward long-term value: whether the content plays a role in a real-world decision-making scenario. This explains a common phenomenon: some content on Red does not make much noise during the initial recommendation phase, yet remains visible over a long period. Its path is often not a one-time explosion, but a gradual entry into search and comparison scenarios, being repeatedly invoked as an "experience sample." Consequently, the lifecycle of such content is significantly longer than content centered on emotional stimulation.

Under this structure, creator credit is formed differently than on other platforms. Red does not particularly emphasize charisma or dramatic storytelling; instead, it cares about a type of predictability: when a user sees you again regarding a similar problem, do you still provide a judgment that is logically consistent, clearly conditioned, and not self-contradictory? Credit is not established once; it accumulates gradually across multiple similar situations. Once this stability is recognized, the account often enters a relatively smooth but continuous state of distribution, making it less likely to be erased entirely because a single post failed.

Accordingly, "strong stimuli" such as clickbait titles, emotional polarization, and the creation of antagonism are not entirely ineffective on Red, but they find it harder to enter long-term, invokable communication paths. The reason is not a platform-level rejection of emotion, but that this content is difficult to use for real-world judgment: it lacks traceable prerequisites, conditions, and consequences, and it is difficult to match with specific questions in a search scenario. Once removed from the immediate context, its utility drops rapidly, and distribution fades accordingly. From the perspective of the overall mechanism, communication on Red does not revolve around "being seen," but around "being used." It does not reward performance for its own sake, nor does it stably reward stance or volume; it favors content that provides a relatively reliable reference for judgment when others are hesitating, comparing, or deciding. In other words, this platform ultimately rewards not "making people remember you," but "allowing people to find you when they need to make a judgment."